Predictive Algorithms and Juvenile Delinquency in Kinshasa: Emerging Challenges and Ethical Implications

This paper evaluates a sample of 444 reported incidents of urban crime in Kinshasa and identifies important trends that can be used to predict juvenile delinquency using predictive modeling. The proportion of incidents is 44.6% assault (198 cases), 39.2% theft (174 cases), and 16.2% vandalism (72 cases), so 48.9% of the incidents occurred in conditions, whereas 48.9% of the assaults occur during the night or evening hours (Jonas et al., 2022; Hunt et al., 2020). Incidents are clustered in municipalities like Limete (14.6% - 65 cases), Ngaliema (9.5%), and Mont-Ngafula (8.1%), frequently in areas remote from law enforcement (46.8%), and with lengthy police response time (median 20 minutes overall, over 75 minutes in distant areas) (Milaninia, 2020; Duursma & Karlsrud, 2019; Majigo, 2023). These spatiotemporal and environmental determinants point to variables that can be used to predict risks using machine learning in urban African settings (Khosa et al., 2024; Ndikumana et al., 2025). The value of the contribution of this study is that it reveals the viability of applying the application of predictive algorithms to the specifics of delinquency data in Kinshasa to identify the hotspots and allocate resources proactively, and critically evaluates the emerging issues associated with the application of predictive algorithms in resource-limited environments (Mancuso & Corselli, 2023; Tolan et al., 2019; Almasoud & Idowu, 2025). The article offers a conceptual roadmap to ethical artificial intelligence application in juvenile justice in urban African settings by promoting a change in the framework where pure risk prediction would be replaced by equity-oriented interventions, which would promote social justice and youth rehabilitation over stigmatization (Berk, 2019; Keddell, 2019; Barabas et al., 2018; Kurumalla, 2025). This model remains consistent with the prevailing demands for transparent, proportional, and rights-oriented predictive tools to prevent harm and improve preventive outcomes (Modise, 2024; Stevenson & Slobogin, 2018; Oswald et al., 2018).